Distributed Inference and Fine-tuning of Large Language Models Over The Internet